Using microwave emissivities for the estimation
of global land surface heat fluxes
Carlos Jimenez1, Catherine Prigent1, Filipe Aires2
1LERMA,
2LMD,
<carlos.jimenez@obspm.fr>
Observatoire de Paris, France
IPSL, Université Paris VI, France
2nd WS on Remote Sensing and Modelling of Surface Properties, Toulouse,
6/2009
Motivation
• Estimating land surface heat fluxes is challenging as the fluxes do not have a
unique signature that can be remotely or directly detected. Some possibilities:
[e.g. monthly latent fluxes August 93]
1.Motivation
2.Satellite
and fluxes
3.Linking
satellite
and fluxes
(a) using observations to infer the
properties of the atmosphere and
surface needed to derive the fluxes
by physically based formulations
e.g. (Fisher J., 2007, Rem. Sens.Envir.)
(b) using observations to force
‘complex’ land surface model,
e.g GSWP-2
(Dirmeyer P. (2006), BAMS)
4.Applications
(c) assimilating observations into
a coupled land-atmosphere model
e.g NCEP reanalysis
(Kalnay. E. (1996), BAMS)
<carlos.jimenez@obspm.fr>
2
Motivation
e.g. latent flux annual climatologies (1993-1995) for different ecosystems
1.Motivation
2.Satellite
and fluxes
3.Linking
satellite
and fluxes
4.Applications
• a large spread of values is observed when comparing the
existing global estimates of land surface heat fluxes.
<carlos.jimenez@obspm.fr>
3
Motivation
• Several global data analysis activities are conducted in the frame of
GEWEX to complete the description of the energy and water cycle.
• Most products are now being worked on (clouds, aerosols, radiative
fluxes, precipitation, ocean surface turbulent fluxes, water vapour,
temperature, and ozone) apart from the land surface heat fluxes.
1.Motivation
2.Satellite
and fluxes
3.Linking
satellite
and fluxes
4.Applications
The LandFlux initiative of the GEWEX Radiation Panel
(GRP):
• Objectives:
to develop the needed capabilities to
produce a global, multi-decadal surface
turbulent flux data product.
• Agenda:
1st workshop in Toulouse, May 2007.
2nd workshop in Melbourne, Aug 2009.
(http://www.gewex.org/projects-GRP.htm)
<carlos.jimenez@obspm.fr>
4
Estimating fluxes: land surface models
• Most global heat flux estimates are coming from coupled/off-line SAVT
schemes with some surface parameters derived from remote sensing data (e.g
LAI), many others from approximate relationships with vegetation, soil type or
climate regime.
e.g. 1993 sensible and latent flux annual mean from two off-line models with similar forcing
ISBA
GSWP-2 B0 baseline run
1.Motivation
2.Satellite
and fluxes
3.Linking
satellite
and fluxes
ORCHIDEE
4.Applications
• Not easy to calibrate/tune the comprehensive parameterizations of
the land models when doing the transition from the local/regional to
the global scale.
<carlos.jimenez@obspm.fr>
5
Estimating fluxes: satellite observations
• There is satellite data with temporal and spatial resolutions compatible with
surface models and with expected sensitivity to the land fluxes.
e.g. monthly mean values for June 93
1.Motivation
2.Satellite
and fluxes
3.Linking
satellite
and fluxes
AVHRR reflectances
[visible and near-IR]
4.Applications
ISCCP skin temperature
[diurnal cycle, thermal infrared]
[Aires et al. (2004), JGR]
<carlos.jimenez@obspm.fr>
6
Estimating fluxes: satellite observations
• There is satellite data with temporal and spatial resolutions compatible with
surface models and with expected sensitivity to the land fluxes.
e.g. monthly mean values for June 93
1.Motivation
2.Satellite
and fluxes
3.Linking
satellite
and fluxes
ERS backscattering
(active microwave)
4.Applications
SSM/I emissivities
[passive microwave]
[Prigent et al. (2006), BAMS]
AVHRR ISCCP
global correlation with NCEP latent fluxes R
<carlos.jimenez@obspm.fr>
0.79
0.76
ERS
SSMI
0.75
0.83
7
Combining observations and model fluxes
• Some of these data are presently not exploited in the calibration/
development of the land surface models, as there is so far no easy
way to integrate these data into the models.
1.Motivation
• Our proposal: to use an statistical model to provide the mapping
between the observations and the land fluxes.
2.Satellite
and fluxes
3.Linking
satellite
and fluxes
Satellite
data
Statistical
model
LSM
fluxes
4.Applications
The relationship between observational data and fluxes is not prescribed, but
derived from the statistical analysis of a global dataset of coincident multifrequency satellite observations and land model fluxes.
<carlos.jimenez@obspm.fr>
8
Combining observations and model fluxes
Example
Applying the statistical modeling with a suite of coincident:
1.Motivation
2.Satellite
and fluxes
3.Linking
satellite
and fluxes
4.Applications
Satellite observations:
Global fluxes:
• ISCCP thermal infrared land skin
• GSWP-2 Global Soil Wetness
temperature [mean value and the
Project exercise
amplitude of the diurnal cycle].
• SSM/I microwave emissivities
[vert/hor polarized at 19, 37 and 85
GHz].
• ERS microwave backscattering
[at 20° and 45° incident angles]
• AVHRR reflectances [visible and
near-infrared].
- 15 LSMs driven in off-line mode using
global meteorological forcing in 19861995.
- using fluxes from a multi-model
analysis (average across the individual
models) and two French participating
models (ISBA and ORCHIDEE).
• NCEP/NCAR reanalysis
- 50 years record frozen global data
assimilation system with a couple landatmosphere scheme.
[ monthly means in 1993-1995, 0.25o x 0.25o]
<carlos.jimenez@obspm.fr>
[Jimenez et al. (2009), JGR]
9
Methodology
Phase 1
The statistical models learn the
global relationships between
observations and fluxes.
1.Motivation
2.Satellite
and fluxes
3.Linking
satellite
and fluxes
4.Applications
[e.g. calibrating with
Feb-May-Aug-Nov 93]
Phase 2
The statistical models map the
observations into fluxes using the
learned global relationships.
[estimating fluxes for
rest of 93-95]
<carlos.jimenez@obspm.fr>
10
Applications
• Producing land heat fluxes
• the observation-driven fluxes are a new product combining the information
from the observations and the land surface model.
e.g. GSWP-mult and NCEP latent fluxes in August 1995
r ~ 0.9, RMSE ~ 25 W/m2
2.Satellite
and fluxes
GSWP
1.Motivation
4.Applications
NCEP
3.Linking
satellite
and fluxes
original
estimated
• Is the new product merging observations and land model outputs a better estimate?
Global evaluation/validation of heat fluxes remains very challenging.
<carlos.jimenez@obspm.fr>
11
Applications
• Model development
• for specific regions and times there may be no consistency between the
LSM fluxes and the statistical model fluxes: this can be used to diagnose
potentially anomalous LSM fluxes.
e.g. averaged fluxes in 2°x2° box around Tapajos Forest [54°W-3°S]
2.Satellite
and fluxes
sensible
1.Motivation
4.Applications
latent
3.Linking
satellite
and fluxes
original
estimated
- in this example the estimated fluxes seem to partition more according
with the expected fluxes that some of the original fluxes.
<carlos.jimenez@obspm.fr>
12
Applications
• Model development
• for specific regions and times there may be no consistency between the
LSM fluxes and the statistical model fluxes: this can be used to diagnose
potentially anomalous LSM fluxes.
e.g. averaged fluxes in 2°x2° box around Tapajos Forest [54°W-3°S]
2.Satellite
and fluxes
sensible
1.Motivation
3.Linking
satellite
and fluxes
original
estimated
4.Applications
- large fluxes at the end of 1995 “corrected” in the
estimated fluxes (fluxes anomaly due an anomaly
in GSWP radiative forcing).
• caution when interpreting the differences, as they can also be related
to problems with the observational data or the statistical model.
<carlos.jimenez@obspm.fr>
13
Applications
• Assimilation
• the observations mapped into state variables by the statistical model
could be integrated into the LSM by standard variational assimilation
schemes.
Cost function to combine information from the observations and the LSM:
1.Motivation
background term
2.Satellite
and fluxes
3.Linking
satellite
and fluxes
4.Applications
LSM
satellite-derived
statistical model error
• there exist techniques to calculate Ri and give more weights to the
statistical model predictions when there are more reliable.
[Aires (2004), JGR]
• as the statistical model was calibrated with the LSM outputs, we force
consistency between LSM and satellite-derived state variables and
minimize problems trying to assimilate exogenous inputs (bias correction,
pdf matching, ….)
<carlos.jimenez@obspm.fr>
14
Summary

Microwave emissivities can be a significant source of information at
the time and spatial resolutions required for the production of a global
climatology of land surface heat fluxes.

We propose an statistical analysis to globally map a dataset of multifrequency observations (including the emissivities) into land surface
model (LSM) heat fluxes.

The satelite-driven fluxes can be considered as a new product that
merges information from the observations and the LSMs.

The methodology can also be used for:
1.Motivation
2.Satellite
and fluxes
3.Linking
satellite
and fluxes
4.Applications
- model development as discrepancies between the original
LSM state variable and the satellite-driven variable can potentially
identify LSM problems.
- including satellite observations into assimilation schemes
minimizing the problems related to assimilating exogenous
inputs.
<carlos.jimenez@obspm.fr>
15
Summary
•
For further details:
[Jimenez, C., C. Prigent, and F. Aires, Toward an estimation of global land
surface heat fluxes from multisatellite observations, J. Geophys. Res., 114,
D06305, 2009.]
[Aires, F., and C. Prigent, Toward a new generation of satellite surface
products?, J. Geophys. Res., 111, D22S10, doi:10.1029/2006JD007362, 2006.]
1.Motivation
2.Satellite
and fluxes
3.Linking
satellite
and fluxes
4.Applications
[Prigent, C., F. Aires, and W. B. Rossow, Land surface microwave emissivities
over the globe for a decade, Bul. Amer. Meteorol. Soc., doi:10.1175/BAMS-8711-1573, 1573-1584, 2006.]
[Aires, F., C. Prigent, W. B. Rossow, Temporal interpolation of global surface
skin temperature diurnal cycle over land under clear and cloudy conditions, J. of
Geophy. Research, 109, D06214, doi:10.1029/2003JD003527, 2004.]
[Aires, F., Neural network uncertainty assessment using Bayesian statistics with
application to remote sensing: 1. Network weights, J. of Geophy. Research,
D10303, doi:10.1029/2003JD004173, 2004.]
Acknowledgments. All data and model teams are acknowledged
by making their observations and model runs available.
<carlos.jimenez@obspm.fr>
16